Proceedings IFKAD 2026

PROCEEDINGS e-books

Proceedings IFKAD 2026

Intelligent Knowledge for Sustainable Organizations

Corvinus University of Budapest, Budapest, Hungary 1-3 July 2026
ISBN: 978-88-96687-21-5
ISSN: 2280-787X

List of Included Articles:

When Knowledge Hiding Enhances Self-Assessment in Multiple Sclerosis

(677-683)

Juan-Gabriel Cegarra-Navarro, Aurora Martínez-Martínez, Scott Erickson, Jorge Cegarra-Sánchez

Abstract

Small variations in how information is framed or selectively presented can significantly shape how individuals interpret and internalize knowledge. Prior research has largely associated knowledge hiding with negative outcomes, partly due to its operationalization as perceived lack of help or misaligned information. In contrast, this study adopts a knowledge regulation perspective, focusing on how selective non-disclosure may shape cognitive processes. Therefore, we adopt a complementary perspective by examining evasive knowledge hiding as a form of knowledge regulation, where information is filtered and sequenced to reduce cognitive overload and prevent the formation of misconceptions.
This study examines how evasive knowledge hiding influences self-assessment among patients with Multiple Sclerosis, a context in which misconceptions and narrow mental models can significantly affect disease management. We argue that learning myopia, understood as the persistence of distorted or incomplete mental models, plays a central role in shaping how patients evaluate their own capabilities and progress. Drawing on survey data from 75 Multiple Sclerosis patients and analysed using Partial Least Squares Structural Equation Modeling (PLS-SEM), the results reveal that evasive knowledge hiding does not directly affect self-assessment. Instead, its impact is fully mediated by learning myopia. Greater evasive knowledge hiding is associated with lower levels of learning myopia, which in turn leads to higher levels of self-assessment. These findings suggest that selective non-disclosure, when enacted as a form of knowledge regulation, may help patients avoid misleading or overly complex information that could distort their understanding.
By challenging the dominant “dark side” view of knowledge hiding, this study highlights its contingent and potentially adaptive role in specific contexts. The findings contribute to knowledge management research by reframing evasive knowledge hiding as a mechanism of knowledge regulation and to healthcare research by showing how managing information flows can support more accurate self-evaluations among patients. Overall, the study uncovers a paradoxical pathway through which knowledge hiding can enhance learning outcomes by reducing cognitive distortions.

AI-Supported Decision-Making in Medical Device Procurement: From Ad Hoc Criteria to Meta-Criteria and Value Configurations

(684-692)

Mattia Barsanti, Salvatore Tallarico, Luisa Pellegrini, Riccardo Dulmin, Simone Lazzini, Marco Praticò

Abstract

Medical device (MD) procurement is a strategic area of non-clinical decision-making in healthcare, as purchasing choices affect not only cost containment but also quality of care, patient safety, and long-term health system sustainability. In competitive tenders, however, evaluation criteria are often highly heterogeneous and device specific. This limits the possibility of systematically comparing procurement decisions across tenders and of understanding the evaluative logics that guide final decisions. Within this context, artificial intelligence (AI) may support procurement not by replacing human judgement, but by structuring fragmented information, identifying recurrent value dimensions, and enabling a more reflexive comparison between AI-generated outputs and observed human decisions.
This study investigates how AI can support decision-making in MD procurement. It focuses on tender documents managed by a regional central purchasing body in Italy and develops a research design structured into four phases: tender document analysis and data extraction; recoding specific-criteria into higher-level meta-criteria; construction of AI-based value configurations; and comparison of these configurations with observed procurement decisions. The paper reports the preliminary results of the first two phases.
The paper reports the preliminary results of the first two phases. As regards the first phase, approximately 500 tenders were identified. After excluding non-MD tenders, price-only procedures, incomplete cases, and non-interpretable documents, the final dataset consisted of 172 tender documents. Preliminary analysis identified three recurring document formats and corresponding extraction schemas.
As regards the second phase, the first results show that device-specific evaluation criteria can be recoded into broader meta-criteria, such as versatility of use, patient safety, operator safety, interoperability, and degree of clinical certainty. These meta-criteria provide the analytical basis for comparing tenders involving different types of MDs and for integrating AI appropriately into procurement decision-making. By preserving traceability to the original criteria while enabling broader comparability, they support AI-based procurement governance and may contribute to accountability and organizational learning by supporting a more explicit justification of procurement decisions over time.

An Integrated Framework for Evaluating Telemedicine Systems in Regional Healthcare Contexts

(693-699)

Lorella Cannavacciuolo, Roberta De Cristofaro, Vincenzo Maione, Cristina Ponsiglione, Simonetta Primario

Abstract

Telemedicine has progressively evolved from a complementary digital service into a structural component of contemporary care models. This transition has transformed telemedicine into a complex intervention whose effectiveness depends not only on technological performance, but also on the dynamic interactions among organizational processes, professional practices, patient engagement, infrastructural conditions, and governance arrangements. Although established multidimensional frameworks, such as Health Technology Assessment (HTA), provide robust methodological foundations for evaluating health technologies, their application at the regional ecosystem level still presents relevant challenges. In particular, traditional assessment approaches often struggle to capture the adaptive nature, contextual variability, stakeholder interdependencies, and systemic trade-offs that characterize telemedicine as a socio-technical system. Against this background, a persistent gap remains between the theoretical recognition of telemedicine complexity and the availability of operational tools capable of monitoring such complexity in an integrated and decision-oriented manner. This study addresses this gap by proposing the Telemedicine Ecosystem Monitoring and Evaluation Framework (TEMEF), an HTA-informed and context-sensitive model designed to support the evaluation of telemedicine systems in regional healthcare contexts. Rather than replacing existing evaluative approaches, TEMEF aims to complement them by translating multidimensional assessment principles into a structured, multi-stakeholder monitoring architecture. Using a Proof of Concept (PoC) methodology, the paper describes the transition from conceptual design to the development of a survey-based evaluation system tailored to four key stakeholder groups: patients and caregivers, physicians, healthcare managers, and technology providers. The framework integrates five main evaluative domains—clinical, organizational, technological, experiential, and economic—allowing system performance to be examined both at an aggregate level and through stakeholder-specific perspectives. To assess the internal coherence of the scoring logic and the interpretability of the analytical outputs, AI-generated synthetic stakeholder profiles were used to conduct a preliminary stress-test of the framework and its analytical dashboard interface. This procedure enabled the exploration of plausible configurations of stakeholder alignment and misalignment, highlighting how apparently positive aggregate performance may conceal critical tensions across domains or actor groups. The resulting model provides policy makers and healthcare decision-makers with an operational framework for interpreting telemedicine not as an isolated technological solution, but as a regional healthcare ecosystem. TEMEF supports evidence-informed governance by making visible systemic trade-offs, implementation bottlenecks, and areas requiring targeted improvement, thereby contributing to the sustainable integration of telemedicine into evolving care delivery models.

The Social Cost of Digital Banking: Branch Desertification and Inequalities in Access to Financial Services

(739-747)

Giuseppe Bruno, Carmela Piccolo, Eduardo Pipicelli

Abstract

In recent years, the banking sector has undergone a profound transformation driven by the widespread diffusion of digital technologies, which has progressively reshaped service delivery models and reduced the role of physical branches. This shift has been accompanied by a significant contraction of branch networks across Europe, raising concerns about its implications for access to financial services, particularly in less advantaged territorial contexts. Against this background, this study investigates the social costs associated with the digitalization of banking services by analyzing how branch network contraction affects spatial accessibility across different territorial contexts. The empirical analysis focuses on a regional case study in Italy, Emilia-Romagna, characterized by a high degree of heterogeneity in socio-economic and digital conditions. Accessibility is assessed through two complementary dimensions: the availability of branches within given distances and the proximity to the nearest facility. The analysis is conducted for three reference years (2010, 2017, and 2024), enabling a detailed comparison of the evolution of accessibility patterns across urban, suburban, and rural areas. The results show that, although branch closures have occurred throughout the region, their impact has been significantly more pronounced in suburban and rural areas, where both availability and proximity indicators reveal a marked deterioration of accessibility conditions. Building on this evidence, the study further explores the relationship between physical accessibility and digital development, showing that the areas most affected by the reduction in physical infrastructure are also characterized by weaker digital conditions. In these contexts, digital channels do not adequately compensate for the loss of physical access, leading to a cumulative effect of disadvantage. Overall, the findings suggest that the ongoing digital transformation of the banking sector entails non-negligible social costs, as it disproportionately affects more vulnerable populations and contributes to widening territorial disparities in access to financial services. The study highlights the need to account for spatial and digital inequalities when designing policies aimed at promoting financial inclusion in the context of digital transition.

How Healthcare Artificial Intelligence Startups Redesign Clinical Processes and Create Value in the Healthcare Ecosystem

(700-709)

Nunzia Carbonara, Davide De Vita, Umberto Panniello, Elisabetta Vurchio

Abstract

Artificial Intelligence (AI) is increasingly recognized as a transformative force in healthcare, with the potential to enhance clinical efficiency, reduce costs, and enable more personalized care pathways. In parallel, AI has stimulated the emergence of numerous startups developing innovative healthcare solutions. However, existing research has largely focused on their technological features and business models, overlooking their broader systemic role. AI startups are entering healthcare ecosystems as new actors, establishing novel relationships with providers, patients, regulators, and technology suppliers, and potentially reshaping structures, coordination mechanisms, and value creation processes. This study addresses this gap by examining how European healthcare AI startups contribute to the redesign of clinical processes and the reconfiguration of value creation within healthcare ecosystems. Building on the ecosystem-as-structure perspective and a role-based view of ecosystem participation, healthcare is conceptualized as a complex system characterized by interdependencies among heterogeneous actors. Within this framework, AI startups are not merely technology providers but actors capable of altering activity flows, redistributing roles, and generating new connections among stakeholders. Adopting a service and ecosystem perspective, the study emphasizes that value does not reside in technological solutions per se but emerges through their use and integration within specific contexts. AI solutions may generate value for patients—through improved accessibility, personalization, and continuity of care—and for providers—through enhanced decision-making and efficiency. At the same time, they may also produce value destruction when they create workflow frictions, usability issues, or misalignment with clinical routines. Focusing on the European context, characterized by strong regulation and data governance constraints, the study employs a mixed-method approach. First, a large dataset of healthcare AI startups is analyzed using a hybrid natural language processing technique to classify firms into core clinical processes (diagnosis, treatment, monitoring/follow-up) and support processes (drug discovery, administrative support, and healthtech infrastructure). Second, user-generated reviews are examined to assess perceived value creation and value destruction across clinical phases and stakeholder groups. Preliminary findings suggest that AI startups play a dual role within healthcare ecosystems, intervening both directly in clinical activities and indirectly through enabling infrastructures and support functions. Their impact extends beyond isolated applications, depending on how they integrate into existing workflows and reshape interactions among ecosystem actors.

A Hybrid RAG Architecture for Knowledge Management in Urban Digital Twins

(710-718)

Giovanni Felle, Antonio Arcadi, Davide Sergio Storelli, Valentina Chetta, Rosaria Lovecchio, Elisa De Bellis

Abstract

Urban Digital Twins (UDTs) have emerged as strategic instruments for urban planning and governance, generating unprecedented volumes of heterogeneous knowledge assets — policy documents, municipal acts, citizen reports, and sensor data streams — that overwhelm traditional Knowledge Management (KM) frameworks. Conventional AI-driven systems struggle to deliver accurate, traceable responses on complex, multi-domain queries, exhibiting persistent hallucination and insufficient contextual grounding: benchmarking studies reveal that advanced LLMs achieve only 34% accuracy on complex queries, standard Retrieval-Augmented Generation (RAG) improves this to merely 44%, and state-of-the-art solutions answer only 63% of questions without hallucination. These limitations are particularly critical in regulated public governance contexts where transparency, provenance, and temporal consistency are fundamental requirements. This paper presents a Hybrid Retrieval-Augmented Generation (H-RAG) architecture for knowledge management in UDT environments, integrating vector-based semantic retrieval with Knowledge Graph-grounded reasoning. Following a Design Science Research (DSR) methodology, we describe the end-to-end architecture of a Knowledge Management System (KMS) comprising six interconnected modules: heterogeneous document ingestion, parsing and normalisation, transformer-based embedding generation, incremental Knowledge Graph construction, hybrid multimodal indexing, and context-fused retrieval and generation. A Knowledge Intake Guard (KIG) ensures semantic consistency and deduplication prior to graph ingestion, preventing knowledge base fragmentation as the document corpus evolves. The KMS adopts a microservices-oriented pattern with asynchronous processing (message broker and task queue) and multimodal storage (object store, vector database, graph database, and search engine), enabling scalable ingestion of structured and unstructured municipal document corpora. Unlike batch-oriented graph approaches, the proposed system supports incremental knowledge enrichment without requiring periodic full reconstruction of the knowledge base. An illustrative case study based on a municipal traffic ordinance demonstrates how the architecture supports cross-silo knowledge integration, serving both citizens submitting natural-language queries and technical operators requiring entity-centric relational retrieval. Benchmarking evidence from the literature confirms substantial accuracy improvements in Knowledge Graph-augmented retrieval settings, with a corresponding significant reduction in hallucinations. The proposed KMS is developed within the AVANT IPCEI-CIS project and contributes: (i) a conceptual framework for hybrid knowledge retrieval in dynamic urban environments; (ii) a replicable DSR-grounded design approach applicable to public administration KM; and (iii) architectural guidelines and design principles for integrating governance-oriented KM capabilities into multi-silo UDT contexts.

Frankfurt Next Generation: Envisioning Urban Futures by Aligning Stakeholders and Impact Holders with Multimodal Participation Formats

(719-729)

Jörg Rainer Noennig, Fernando Montano, Anja Jannack, Sebastian Wiesenhütter

Abstract

Public participation has become an established instrument of responsive urban development, yet large-scale processes often struggle to combine broad accessibility, methodological diversity, and a clear pathway from citizen input to actionable outcomes. This paper examines “Frankfurt Next Generation” (FNG), a one-year urban future-visioning and participation process initiated by the Polytechnische Gesellschaft Frankfurt (PTG) and scientifically supported by HafenCity University Hamburg and TU Dresden. The project sought to engage a wide spectrum of Frankfurt’s urban society while, in parallel, activating influential organisational actors—termed “Impact Holders”—whose social, economic, and symbolic capital could support the implementation of emerging initiatives.
Methodologically, FNG combined multiple digital and on-site participation formats within an integrated process choreography. Building on an extended Double Diamond approach, the project developed a “Diamond Cascade” that progressively transformed initial impulses into urban future visions and, ultimately, concrete project ideas. Digital participation, a multilingual public placation campaign, qualitative coding, geospatial analysis, local co-creation workshops, and a final “Future Festival” were connected through defined convergence points between the broad stakeholder track and the Impact Holder track. The online campaign generated a substantial body of citizen input, including self-formulated future visions. In a subsequent qualitative analysis, the contributions were coded, grouped into thematic complexes, and ultimately structured into 14 “Vision Containers”, preserving the diversity of citizen perspectives rather than reducing them to a single overarching vision.
The process culminated in six “Lighthouse Projects”, including initiatives on climate adaptation, education, culture, health, and new spaces of opportunity. The findings indicate that FNG provides a proof-of-method for large-scale multimodal participation by combining quantitative reach with targeted qualitative engagement and by linking broad civic input to actors capable of sustaining follow-up activities. At the same time, the project revealed limitations in processing extensive qualitative datasets and in moving beyond activation and consultation toward deeper forms of co-design and conflict mediation. The paper therefore argues that FNG is best understood not as a standardized participation model, but as a replicable and adaptable process architecture for orchestrating heterogeneous participation formats, actors, and outputs in urban future-making.

Path Development and the Twin Transition in Europe

(730-738)

Vincenzo Maione, Cristina Ponsiglione, Simonetta Primario

Abstract

A central challenge in regional innovation and sustainability transitions research is to explain why the twin innovation transition — understood as the joint advancement of green and digital technological specialisation — emerges only in some regions and only at specific moments in time. While Evolutionary Economic Geography has offered powerful concepts to analyse regional path development, much of this literature has focused on the emergence of new growth trajectories, paying less attention to situations in which path-development potential stabilises, remains latent, episodic, or insufficiently converted into transformative innovation outcomes.
This paper addresses this gap by developing a configurational panel framework that links regional path development mechanisms to twin innovation outcomes across European NUTS-2 regions over the period 2007–2019. Drawing on regional path development theory and sustainability transitions research, we argue that twin innovation is not the result of a single enabling factor, but of place-specific combinations of structural assets, innovation capabilities, external linkages, policy support, and institutional conditions. At the same time, the absence of twin innovation should not be interpreted simply as a lack of capabilities, but may reflect incomplete conversion processes, weak cross-domain coupling, or path-development dynamics that remain temporally unstable.
Methodologically, the paper applies panel fuzzy-set Qualitative Comparative Analysis to distinguish between cross-sectional variation across regions and longitudinal variation within regions. This allows us to assess not only which configurations are sufficient for twin innovation, but also whether they display temporal stability or instead appear as episodic alignments. The findings show that regional innovation performance constitutes an important baseline condition across the identified configurations, but is not sufficient on its own to generate twin innovation outcomes. Moreover, the mechanisms most commonly associated with cumulative regional development, such as extension and branching, do not emerge as sufficient pathways to twin innovation. Instead, the results point to a limited number of highly place-specific configurations, often concentrated in Northern and Western European regions, whose empirical relevance is frequently episodic rather than stable over time.
The paper contributes to the literature in two ways. Conceptually, it refines the link between regional path development and sustainability transitions by distinguishing between path-development potential, path activation, and path realisation. Empirically, it provides systematic panel evidence on the spatial selectivity and temporal fragility of twin innovation trajectories in Europe, with implications for place-based innovation policies aimed at supporting not only regional capabilities, but also their effective conversion into green-digital transformation.

Assessing VSME Alignment in Voluntary Sustainability Reporting by SMEs: Evidence from the Italian Chemical Sector

(748-755)

Manuel Cavola, Patrice De Micco, Niccolò Parissi

Abstract

Small and medium-sized enterprises (SMEs) play a central role in the European transition towards a more sustainable economy, yet their sustainability reporting practices remain less developed and less standardised than those of large firms. This issue is particularly relevant in the context of the Corporate Sustainability Reporting Directive (CSRD), the European Sustainability Reporting Standards (ESRS), and the increasing sustainability-related information demands that SMEs face from large companies, financial institutions, and other stakeholders. In response to proportionality concerns, the EFRAG Voluntary Sustainability Reporting Standard for non-listed SMEs (VSME) provides a simplified framework intended to support voluntary sustainability disclosure by SMEs. However, empirical evidence on the extent to which SME sustainability reports are aligned with the VSME remains limited.
This paper assesses the alignment of voluntary sustainability reporting by SMEs operating in the Italian chemical sector with VSME disclosure requirements. The empirical analysis focuses on firms classified under the C20 division of the NACE industrial classification. Starting from a population of Italian chemical firms identified through the AIDA database, the study selects SMEs that had published a sustainability report or an equivalent sustainability-related disclosure document. A stratified sampling strategy was then adopted, and the final analytical dataset was obtained after a quality-control procedure on document extraction.
The analysis applies an AI-assisted content analysis protocol based on a VSME coding grid composed of 20 disclosure items, covering the Basic and Comprehensive Modules and four thematic areas: General information, Environment, Social, and Governance. Each firm-item pair was assessed through a three-level scoring scale distinguishing between no alignment, partial alignment, and full substantive alignment with the VSME requirements.
The results show a moderate and uneven degree of alignment with the VSME. Full and partial alignment are both relatively frequent, while non-coverage remains substantial but not dominant. At firm level, the findings indicate relevant heterogeneity across companies. General information records the highest alignment, followed by Social and Environment, whereas Governance is the weakest area. At item level, firms perform better on business-model descriptions, sustainability initiatives, and human-rights policies, while weaker alignment emerges for biodiversity, pollution, climate-transition targets, corruption-related sanctions, and sensitive-sector revenues. Overall, the findings suggest that Italian chemical SMEs are more advanced in qualitative and narrative sustainability reporting than in systematic, quantitative, and VSME-specific disclosure.

Artificial Intelligence and Labour Law: The Employer’s Obligation to Reskill Workers in the Digital Era

(756-765)

Antonio Capasso, Annabella Lattarulo

Abstract

Artificial Intelligence is driving unprecedented structural reconfiguration of labour markets, yet regulatory frameworks governing employment relationships remain anchored to twentieth-century industrial paradigms. This paper addresses a critical research gap: the absence of coherent legal frameworks allocating responsibility for workers’ reskilling among employers, workers, and public institutions in contexts of AI-driven technological transformation. Drawing on comparative legal analysis and organizational learning theory, this study examines how jurisdictions—specifically Italy, the European Union, the United States, and selected developing countries—approach the regulatory challenge of employer-mandated reskilling obligations. The paper proposes a multidimensional regulatory framework articulated along five core dimensions: legal recognition of reskilling as an employment relationship component, cost-sharing mechanisms, algorithmic oversight, collective bargaining strengthening, and corporate due diligence integration. Through this framework, the paper contributes theoretically to understanding the intersection of technological innovation, labour law, and organizational learning, while offering practical recommendations for legislators and international organizations navigating AI governance in employment contexts. The analysis reveals that while the European Union has advanced more structured approaches through the AI Act, significant implementation inconsistencies persist, and integrations with labour law remain underdeveloped. Conversely, the United States exhibits fragmentation across state-level frameworks, while developing countries face structural barriers to coherent reskilling policies. The paper demonstrates that sustainable AI governance requires reconceptualizing the social contract underlying employment relationships, ensuring that technological transitions develop in alignment with adequate worker protection and organizational learning capacity.

FedHR5.0: A Human‑Centric Federated Learning Framework for Organisational Learning and Workforce Competency Development in Healthcare

(766-774)

Fabio Liberti

Abstract

Healthcare organisations face simultaneous pressure to digitalise their workforce, redesign learning processes and adopt artificial intelligence in support of human resource decisions, while protecting sensitive employee data and respecting professional dignity — pressures more acute in healthcare than in other knowledge-intensive sectors, and bearing on clinical and administrative cadres alike.
Three research streams have approached this problem in isolation: federated learning in healthcare, privacy-preserving HR analytics, and organisational learning theory engaging with AI as a partner in knowledge manage-ment. Federated learning has predominantly addressed clinical data rather than workforce competency data; HR analytics has not been developed at the intersection with federated infrastructures; organisational learning theory has rarely engaged with the constraints that federated, privacy-preserving learning imposes. We position our contribution explicitly at the three-way intersection.
We introduce FedHR5.0, a conceptual framework for privacy-preserving organisational learning and workforce competency development in healthcare. FedHR5.0 combines federated learning, (ε, δ)-differential privacy and transparent governance with a five-module architecture — trust-based, adaptive learning, emotionally-intelligent assessment, ethical AI, and collaborative knowledge exchange. The framework extends the SECI dynamics of Nonaka and Takeuchi toward a generative-AI-aware GRAI cycle, in which tacit–explicit transitions are mediated by an explicit privacy budget that doubles as a trust-building instrument. The artefact is positioned as a Design Science Research improvement-type contribution.
Empirical validation is scoped as future work and given a concrete pathway: a multi-hospital pilot, four opera-tionally measurable evaluation axes (model accuracy, effective privacy budget, organisational readiness, worker participation) and a three-phase iterative protocol grounded in Design Science Research.
For HR practice, the framework repositions learning and development as a strategic driver of human–AI collabora-tion and treats workforce participation as a structural design requirement, applicable to clinical and administrative cadres alike. For research, it offers a starting point for cumulative design knowledge at the emerging intersection of federated learning, organisational learning and human-centric HR analytics.

Research Methods for Sustainable Organizations Using Intelligent Knowledge

(775-780)

Pierluigi Mascaro

Abstract

The transition toward sustainable organizations has intensified the need for advanced knowledge systems capable of addressing complex economic, environmental, and social challenges. In this context, intelligent knowledge—defined as the integration of human expertise, organizational learning, and digital technologies such as artificial intelligence (AI) and data analytics—has emerged as a critical enabler of informed decision-making and long-term organizational resilience. This paper examines the research methods most suitable for investigating sustainable organizations through the lens of intelligent knowledge, with particular attention to their ability to capture complexity, dynamism, and socio-technical interactions.
The study adopts an integrative methodological perspective, combining insights from quantitative, qualitative, mixed-methods, design science, and participatory research approaches. Quantitative methods support the analysis of measurable relationships between knowledge capabilities and sustainability performance, while qualitative approaches provide in-depth understanding of organizational processes, cultural dynamics, and meaning-making practices. Mixed-methods designs enable the triangulation of findings, enhancing both validity and explanatory power. In addition, design science and action research approaches contribute to bridging theory and practice by supporting the development and evaluation of intelligent knowledge systems in real organizational contexts.
A central argument of the paper is that no single methodological approach is sufficient to address the multifaceted nature of sustainability challenges. Instead, the research highlights the necessity of methodological pluralism, emphasizing longitudinal and process-oriented designs, multi-level analysis, and the integration of ethical considerations related to AI and data use. The paper also underscores the importance of reflexivity and interdisciplinary collaboration in developing robust and context-sensitive research frameworks.
By providing a comprehensive overview of methodological approaches and their implications, this study contributes to the advancement of sustainability research and offers a coherent framework for analyzing how intelligent knowledge can support sustainable organizational transformation. The proposed methodological orientation is both theoretically grounded and practically relevant, enabling researchers and practitioners to better understand and navigate the complexities of sustainability in an increasingly data-driven world.

Artificial Intelligence and Soft Skills Development in Organizations: A Systematic Literature Review

(781-790)

Antonio Lorena, Anna M. Correale, Alessia Talarico, Rocco Reina

Abstract

The rapid integration of Artificial Intelligence (AI) into organizational contexts is progressively reshaping how knowledge and skills are developed and managed (Kaplan & Haenlein, 2019; Dwivedi et al., 2021). While AI-driven automation has transformed processes and decision-making, its implications for people’s learning, particularly about soft skills, remain an area of growing scholarly and managerial interest. The current evidence base is still fragmented, and recent contributions have explicitly called for systematic reviews to clarify how AI supports (or constrains) soft skills (Deming, 2017; Heckman & Kautz, 2012; Balcar, 2016). Against this background, the present study adopts a Systematic Literature Review (SLR) methodology guided by the PRISMA 2020 framework (Page et al., 2021; Tranfield et al., 2003). The review focuses on peer-reviewed articles published in English, covering all available years up to February 2026. This inclusive temporal scope was adopted to provide a comprehensive view of the evolution of research on AI and soft skills development, while also incorporating the most recent contributions in the field. The search is conducted across three important academic databases: Scopus, Web of Science, and ProQuest, using the core query “Artificial Intelligence” AND “soft skills”, adapted through wildcards to reflect database-specific syntax. The selection procedure follows a multi-stage screening protocol and includes a quality appraisal of the eligible studies. The review is guided by the following research question: How is Artificial Intelligence shaping the landscape of soft skills development in organizations? The synthesis is likely to highlight recurring patterns, and at the same time, expected to systematize evidence on relevant challenges such as algorithmic bias, data privacy, and the risk of dehumanizing learning contexts (Morandini et al., 2023). Overall, the study contributes by mapping the existing literature in a structured and transparent manner, while also offering actionable implications for professionals, educators, and policymakers seeking to harness AI in support of a future-ready workforce.

Explainable Artificial Intelligence as an Organisational Learning Lever: A Theoretical-Methodological Framework for Professional Upskilling in AI-Augmented Workplaces

(791-799)

Raimondo Fanale

Abstract

The integration of Artificial Intelligence (AI) into professional practice is reshaping how knowledge is created, validated, and transmitted within organisations. While AI-driven decision support systems offer considerable analytical power, many rely on deep learning architectures that operate as black boxes, producing outputs that are difficult for human users to understand or critically evaluate. Explainable Artificial Intelligence (XAI) has emerged as a response to this opacity, yet its role has been framed predominantly in terms of technical transparency and user trust, underexploring its potential as a lever for professional development and organisational learning. This paper makes a theoretical-methodological contribution by introducing the XAI-OLL (Explainable Artificial Intelligence as an Organisational Learning Lever) framework, which reconceptualises XAI-generated explanations not merely as transparency mechanisms but as cognitive and organisational artefacts capable of activating processes of professional upskilling and knowledge integration. The framework proposes a sequential cognitive mechanism in which exposure to XAI explanations stimulates reflective reasoning — the deliberate comparison between algorithmic outputs and one’s own judgment — which in turn fosters interpretive competence, defined as the capacity to critically engage with AI-generated evidence. This pathway is conditioned by explanation quality: imperfect or misleading explanations can attenuate or reverse the effect on calibrated trust. Interpretive competence enables calibrated trust, a proportionate form of reliance grounded in evidence quality rather than automation deference. Calibrated trust, in turn, facilitates the integration of AI-derived insights into shared professional knowledge, supporting organisational learning at the collective level. The XAI-OLL framework generates five testable propositions, operationalised through a pilot quasi-experimental protocol in AI-supported histopathology decision-making. The paper contributes to knowledge management, organisational learning, and HR transformation literature by positioning AI interpretive literacy — the competence to engage critically with XAI outputs — as a new dimension of professional capability in AI-augmented workplaces. Practical implications address HR professionals as active designers of XAI-enabled upskilling pathways, governance conditions, and organisational learning infrastructures.

Metrics and Value of Intelligent Knowledge: How Organizations Turn AI Options into Decision Capability, and Become More Sustainable

(800-806)

Harald Grundner

Abstract

AI, copilots and autonomous agents have drastically lowered the cost of producing information and alternatives. Yet many organizations do not become faster or more effective in decision-making. Instead, they experience option overload, longer alignment cycles and persistent uncertainty at the decisive moment. This paper argues that the bottleneck has shifted from generating knowledge artefacts to building organizational decision capability.
The practitioner message is simple: value is always two-sided: customer benefit and economic gain for the organization. Intelligent knowledge is therefore only valuable when it enables better decisions that increase customer benefit and improve value capture. Sustainability in day-to-day management becomes tangible when results and artefacts are reusable: a product (material or immaterial, such as a copilot, agent or prompt) becomes sustainable only if its use repeatedly creates value for users and customers.
We define intelligent knowledge pragmatically as knowledge that makes decisions actionable, accountable and auditable. To measure its value, we propose a metric architecture with three layers. Outcome metrics ask whether decisions improved results and reduced rework and waste. Flow metrics ask whether the right knowledge reached decisions in time and whether it became reusable as a standard or building block. Leadership and governance metrics ask whether decisions are owned, traceable and trustworthy, including disciplined use of AI.
To operationalize the framework, we embed it in a recurring rhythm: Learn – Measure – Build/Test – Decide. Each leaf ends with a decision gate. New insights can force a sharp return and re-entry, ensuring learning is woven back into the process. Each cycle yields two outputs: a decision-ready set for the current decision episode, and a process result that updates knowledge assets (standards, templates, prompts, playbooks) for continuous improvement. An illustrative scenario shows how organizations can transform AI-driven option abundance into calmer, faster and more sustainable decisions by making value criteria explicit and by measuring both decision impact and reusability.

From Quantitative ESG Data to Qualitative Social Insights: A Low-Code AI Agent for Decision Support

(807-814)

Riccardo Censi, Paola Campana, Fulvio Schettino, Chiara De Pucchio, Valentina Ferrari

Abstract

This paper explores the use of Large Language Model (LLM)-based agents as tools for supporting knowledge creation from Environmental, Social and Governance (ESG) data, with a specific focus on the social dimension. Despite the growing availability of ESG data, organisations still face substantial challenges in transforming quantitative indicators into actionable insights that can effectively support strategic decision-making. Current reporting tools mainly provide descriptive outputs, while more advanced AI-based solutions often require technical expertise that limits their practical adoption. This study addresses that gap by proposing an accessible, low-cost decision support system built on open-source technologies. The research adopts a Design Science Research (DSR) approach and develops a conversational ESG advisory agent built on Langflow, an open-source low-code platform, and Llama 4 Scout 17B, a large language model served free of charge via Groq. The agent receives structured ESG indicators through a guided conversational interface, computes relevant metrics, compares them against sector benchmarks derived from principal component analysis and cluster analysis of real energy-sector data, and produces a qualitative assessment report including an improvement roadmap aligned with Global Reporting Initiative (GRI) 401-406, European Sustainability Reporting Standards (ESRS) S1-S2, and Sustainable Development Goals (SDGs) 3, 5, and 10. Evaluation on a synthetic dataset of 40 fictitious energy companies shows that the agent achieves 98.1% value accuracy and 93.2% judgment accuracy on eight of the eleven variables tested. The remaining three variables, specifically the Gender Pay Gap and the two Cultural Diversity indicators, show significantly lower accuracy ranging from 33% to 45%, attributable to the model’s limited capacity to handle competing instructions and maintain parallel reasoning tracks. Through iterative DSR testing, the study finds that restructuring the input to eliminate ambiguity is a more reliable corrective strategy than refining the prompt, suggesting a broader design principle for low-code AI systems based on mid-sized models. The paper contributes to the emerging literature on AI-driven sustainability analytics by proposing and evaluating a replicable, low-cost system that bridges the gap between quantitative ESG reporting and qualitative organisational insight. From a practical standpoint, the findings demonstrate that accessible AI architectures can support meaningful ESG knowledge creation even in resource-constrained organisational settings, acting as a cognitive support tool for human decision-makers rather than replacing them.

Contextual–RFM (C-RFM): a Multimodal Framework for the Fusion of Transactional and Semantic Data Through LLMs for Interactive Customer Segmentation

(815-823)

Davide Madonna, Simone Malacaria, Andrea De Mauro

Abstract

This study examines the evolution of customer segmentation from rule-based systems to multimodal AI architectures, emphasizing the necessity for transparent and context-aware CRM strategies. Despite the widespread adoption of RFM models, organizations often struggle to synthesize transactional metrics with multifaceted qualitative feedback, creating a numerical paradox that this research addresses by capturing the authentic voice of the customer within high-dimensional analytical spaces. This paper presents the Contextual-RFM (C-RFM) framework, an innovative multimodal architecture aimed at bridging the representational divergence between quantitative metrics and qualitative feedback.
The research operationalizes a comprehensive five-phase pipeline to ensure methodological transparency and replicability: (1) establishment of a statistical baseline, (2) narrative synthesis for tabular serialization, (3) multimodal vectorization utilizing advanced embeddings, (4) semantic analysis through Latent Dirichlet Allocation (LDA) and (5) implementation of the C-RFM Expert chatbot. This entire workflow is implemented using KNIME Analytics and Langflow, providing a robust and replicable process for both scholars and practitioners. A key distinction of the framework is its early-fusion approach. Unlike traditional late-fusion models that integrate data streams only at the final analysis stage, early-fusion retains critical cross-modal correlations during the initial processing phase, effectively addressing the semantic opacity common in traditional numerical-only datasets.
Empirical findings indicate that the C-RFM framework adeptly organizes linguistic complexity, evidenced by a significant improvement in structural validity when transitioning from purely transactional environments to topic-validated contexts. Moreover, this study introduces the Discrepancy Detection Rate (DDR) as a novel performance metric to quantify the misalignment between transactional behaviour and qualitative feedback, thereby identifying critical churn risks that conventional quantitative analyses often overlook.
The primary contribution of this research lies in the synergy achieved between human expertise and Artificial Intelligence. By deploying dynamic persona chatbots grounded in Retrieval-Augmented Generation (RAG), the framework facilitates a shift from opaque analytical outputs to actionable, customized recovery strategies. This glass-box methodology transforms intricate numerical complexities into auditable business narratives, empowering decision-makers to transcend statistical constructs and foster a deeper, dialogue-based understanding of consumers. The C-RFM framework represents a significant advancement in human-centred AI, promoting strategic marketing intelligence that is not only transparent but also finely tuned to the authentic voices of real-world consumer archetypes, thereby paving the way for the next generation of customer-centric strategies.

Retrieval-Augmented Generation in Business Applications: A Systematic Literature Review and Taxonomy

(824-832)

Beren Çay, Mario Miozza, Andrea De Mauro

Abstract

Retrieval-Augmented Generation (RAG) has become one of the most visible mechanisms for grounding large language model outputs in organizational knowledge sources. While the technical literature on RAG architectures is substantial, the evidence on how RAG is deployed inside organizations, and on the organizational consequences of those deployments, has not been systematically synthesized. This paper addresses that gap by treating RAG as a hybrid knowledge capability pairing a generative model with curated organizational knowledge and human oversight. The study reports a systematic literature review of peer-reviewed work on RAG in business and organizational contexts published between 2023 and 23 February 2026, covering Scopus and Web of Science in line with the PRISMA 2020 reporting framework. From 727 records after database-side filtering, 513 were screened and 41 retained. The studies were synthesized through the empirical-to-conceptual taxonomy development method of Nickerson, Varshney and Muntermann (2013) into a three-dimensional framework linking business application contexts, RAG architectures, and reported organizational consequences. Three findings stand out. First, adoption concentrates in knowledge-intensive support functions: enterprise knowledge retrieval, document intelligence, and decision support together account for around seventy percent of the corpus. Second, Classic RAG remains the most frequent single configuration, while Advanced and Agentic variants together appear in 18 studies and concentrate in deployments with more demanding task structures. Third, the most consistent finding is not a benefit but a limitation: residual output quality issues appear in 26 of 41 studies and recur across every architectural category and every application context, while data-quality dependence is the most frequent implementation challenge. The paper contributes a business-oriented taxonomy empirically distinct from architecture-first surveys, positioning RAG as a socio-technical capability whose organizational value depends on the surrounding data, governance, and oversight infrastructure. For practitioners, it supports deliberate decisions about where RAG delivers value, which architecture fits which task, and which conditions enable successful deployment.

The Digital Platform as Cognitive Infrastructure: Investigating Human-AI Synergy and Algorithmic Governance in Investment Management

(833-841)

Paola Demartini, Alessandro Salvatori

Abstract

Over the past two decades, the financial sector has undergone a profound transformation, firmly positioning information as its primary competitive asset and placing Artificial Intelligence (AI) at the core of this shift. By interpreting diverse data volumes at speeds beyond human capacity, AI serves as a critical supportive technology. Within this landscape, digital platforms like BlackRock’s Aladdin have emerged as essential cognitive infrastructures functioning as comprehensive operating systems for modern finance. These platforms seamlessly integrate the entire investment lifecycle into a unified data layer, reshaping organizational roles and business models by facilitating a vital interaction between human insight and algorithmic intelligence.
Despite AI’s well-documented technical capabilities, a critical gap remains in understanding the socio-technical dynamics of this human-AI synergy within institutional settings. Current literature predominantly focuses on the predictive accuracy of “black box” models or macro-level regulatory frameworks, overlooking how professionals navigate daily trade-offs between algorithmic performance and explainability. Furthermore, there is limited empirical exploration of the modern financial analyst’s “hybrid profile,” which requires combining computational literacy with ethical sensitivity within established platform constraints.
To address this gap, this study investigates how digital investment platforms facilitate human-AI synergy while managing structural risks of algorithmic opacity and bias. Employing a qualitative, exploratory approach, the research utilizes semi-structured interviews with senior European Union professionals involved in investment products, digital financial advisory, platform delivery, and risk-related decision-making. The investigation focuses on integrating AI-driven insights into fundamental analysis, perceptions of algorithmic opacity, the EU AI Act’s impact on governance, and platforms’ role in standardizing information.
Preliminary findings indicate that rather than replacing human experience, AI fundamentally reconfigures it, shifting professional judgment toward problem definition, exception management, and robustness testing. Integrated platforms provide a “common data language” that aligns internal functions, significantly reducing manual errors and operational risk. However, a persistent tension exists between maximizing predictive power and maintaining the transparency demanded by the EU AI Act, which frames AI as a corporate governance issue rather than mere efficiency. Ultimately, the study highlights a “new responsibility pact” requiring auditable decision-making, while warning that “algorithmic convergence” through dominant platforms could dangerously amplify market volatility. Sustainable innovation requires these cognitive infrastructures to be supported by robust data governance.

Perturbation-First Robustness Evaluation of Automated Short Answer Grading

(842-849)

Ferdinando Sasso, Andrea De Mauro

Abstract

Automated Short Answer Grading (ASAG) is increasingly deployed in high-stakes educational contexts, yet the dominant evaluation paradigm focuses primarily on agreement with human raters, overlooking robustness as a complementary quality dimension. Robustness, defined as the consistency of grading behaviour under controlled, meaning-aware input transformations, bears directly on construct validity, the degree to which an assessment measures what it claims to measure, and conditions the trustworthiness of automated decisions in education.
We propose a perturbation-first evaluation framework that assesses robustness across three validity-theoretic dimensions: invariance to construct-irrelevant variation, sensitivity to meaning-altering changes, and resistance to adversarial gaming. The framework combines rule-based perturbation generators, two-gate quality validation, and self-referential metrics that compare the grader against its own output rather than against gold labels, isolating robustness from accuracy. A dual-protocol design contrasts within-question evaluation, the optimistic scenario typically reported in the literature, with cross-question evaluation, the realistic deployment scenario, capturing how trained graders degrade under distribution shift.
We apply the framework to a benchmark ASAG dataset of science short answers and compare a trained machine-learning baseline with zero-shot Large Language Model (LLM) graders evaluated at two prompting levels, with and without access to a reference answer. Three findings emerge. First, trained graders exhibit substantial robustness degradation under realistic cross-question evaluation, suggesting that standard within-question benchmarks overestimate operational reliability. Second, zero-shot LLM graders exhibit stronger robustness on invariance and sensitivity dimensions while remaining comparatively less vulnerable to adversarial manipulation, although residual sensitivity failures display a stable hierarchy across perturbation types. Finally, providing reference answers introduces a subtle trade-off, with improved consistency on meaning-preserving variations but increased susceptibility to keyword-based gaming, which is invisible to single-metric evaluation.
We argue that multi-dimensional robustness evaluation should complement accuracy-based metrics in ASAG systems intended for high-stakes educational use, and we release an open-source framework that operationalises the approach and is freely available to the research community.